Thomas A Gerds

IT University of Copenhagen, København, Capital Region, Denmark

Are you Thomas A Gerds?

Claim your profile

Publications (124)382.96 Total impact

  • [Show abstract] [Hide abstract]
    ABSTRACT: Objectives: A recent alarming finding suggested an increased risk of renal tumors among long-term lithium users. The objectives of the present study were to estimate rates of renal and upper urinary tract tumors (RUT), malignant and benign, among individuals exposed to successive prescriptions for lithium, anticonvulsants, and other psychotropic agents used for bipolar disorder, and among unexposed individuals. Methods: This was a nationwide, population-based longitudinal study including time-specific data from all individuals exposed to lithium (n = 24,272) or anticonvulsants (n = 386,255), all individuals with a diagnosis of bipolar disorder (n = 9,651), and a randomly selected sample of 1,500,000 from the Danish population. The study period was from 1995 to 2012, inclusive. Outcomes were hazard rate ratios (HR) for RUT in three groups: (i) combined malignant and benign, (ii) malignant, and (iii) benign. Analyses were adjusted for the number of prescriptions for lithium/anticonvulsants, antipsychotic agents, antidepressants, and use of all other types of medication; age; gender; employment status; calendar year; and a diagnosis of bipolar disorder. Results: Continued treatment with lithium was not associated with increased rates of RUT [adjusted HR malignant or benign: 0.67-1.18, p (trend) = 0.70; adjusted HR malignant: 0.61-1.34, p (trend) = 0.90; adjusted HR benign: 0.74-1.18, p (trend) = 0.70]. Similarly, continued treatment with anticonvulsants was not associated with increased rates of RUT [adjusted HR malignant or benign: 0.97-1.18, p (trend) = 0.10; adjusted HR malignant: 0.82-1.15, p (trend) = 0.80; adjusted HR benign: 0.94-1.36, p (trend) = 0.20]. The associations were confirmed among the 9,651 patients with a diagnosis of bipolar disorder. Conclusions: Treatment with lithium is not associated with increased rates of RUT.
    Bipolar Disorders 11/2015; DOI:10.1111/bdi.12344 · 4.97 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Importance: Lithium is the main mood stabilizing drug for bipolar disorder. However, it is controversial whether long-term maintenance treatment with lithium or other drugs for bipolar disorder causes chronic kidney disease (CKD). Objective: To compare rates of CKD and in particular rates of end-stage CKD among individuals exposed to successive prescriptions of lithium, anticonvulsants, or other drugs used for bipolar disorder. Design, setting, and participants: This is a Danish nationwide population-based study of 2 cohorts. Cohort 1 comprised a randomly selected sample of 1.5 million individuals among all persons who were registered in Denmark on January 1, 1995, all patients with a diagnosis of a single manic episode or bipolar disorder between January 1, 1994, and December 31, 2012 (n =10 591), and all patients exposed to either lithium (n = 26 731) or anticonvulsants (n=420 959). Cohort 2 included the subgroup of 10 591 patients diagnosed as having bipolar disorder. Main outcomes and measures: Possible CKD, definite CKD, and end-stage CKD (defined as long-term dialysis or renal transplantation). Results: A total of 14 727 (0.8%), 18 762 (1.0%), and 3407 (0.2%) in cohort 1 and 278 (2.6%), 319 (3.0%), and 62 (0.6%) in cohort 2 were diagnosed as having possible, definite, or end-stage CKD, respectively. Based on the total sample and not considering diagnoses, use of lithium was associated with an increased rate of definite CKD (0 prescriptions: hazard ratio [HR] = 1.09, 95% CI, 0.81-1.45; ≥60 prescriptions: HR = 3.65, 95% CI, 2.64-5.05; P for trend < .001) and possible CKD (0 prescriptions: HR = 1.01, 95% CI, 0.79-1.30; ≥60 prescriptions: HR = 2.88, 95% CI, 2.17-3.81; P for trend < .001), whereas use of anticonvulsants, antipsychotics, or antidepressants was not. Neither use of lithium nor use of any other drug class was associated with increasing rates of end-stage CKD. In patients with bipolar disorder, use of lithium was associated with an increased rate of definite CKD (1-2 prescriptions: HR = 0.89, 95% CI, 0.39-2.06; ≥60 prescriptions: HR = 2.54, 95% CI, 1.81-3.57; P for trend < .001) or possible CKD (1-2 prescriptions: HR = 1.26, 95% CI, 0.65-2.43; ≥60 prescriptions, HR = 2.48, 95% CI, 1.80-3.42; P for trend < .001), as was use of anticonvulsants (definite CKD, 1-2 prescriptions: HR = 1.23, 95% CI, 0.76-1.99; ≥60 prescriptions, HR = 2.30, 95% CI, 1.53-3.44; P for trend < .001; possible CKD, 1-2 prescriptions: HR = 1.11, 95% CI, 0.70-1.76; ≥60 prescriptions: HR = 1.97, 95% CI, 1.34-2.90; P for trend < .001). There was no such association with antipsychotics or antidepressants. Also in patients with bipolar disorder, use of lithium was not significantly associated with an increased rate of end-stage CKD, whereas use of anticonvulsants was (1-2 prescriptions, HR = 0 [95% CI, 0.00-infinity]; 30-39 prescriptions: HR = 3.23, 95% CI, 1.26-8.27; ≥60 prescriptions: HR = 2.06, 95% CI, 0.82-5.16; P for trend = .002). Conclusions and relevance: Maintenance treatment with lithium or anticonvulsants as practiced in modern care is associated with an increased rate of CKD. However, use of lithium is not associated with an increased rate of end-stage CKD. The associations between use of medication and CKD may at least partly be attributed to bias.
    JAMA Psychiatry 11/2015; DOI:10.1001/jamapsychiatry.2015.1834 · 12.01 Impact Factor
  • Eva Lauridsen · Thomas Gerds · Jens Ove Andreasen ·
    [Show abstract] [Hide abstract]
    ABSTRACT: AimTo analyze the risk of pulp canal obliteration (PCO), pulp necrosis (PN), repair-related resorption (RRR), infection-related resorption (IRR), ankylosis-related resorption (ARR), marginal bone loss (MBL), and tooth loss (TL) for teeth involved in an alveolar process fracture and to identify possible risk factors.Material and methodA total of 91 patients with 223 traumatized teeth.StatisticsThe risks of PCO, PN, RRR, IRR, ARR, MBL, and TL were analyzed separately for teeth with immature and mature root development using Kaplan–Meier and Aalen–Johansen methods. Possible risk factors for PN (age, fracture in relation to apex, displacement, gingival injury, degree of repositioning, type of splint, duration of splinting, treatment delay, and antibiotics) were analyzed for mature teeth using Cox regression. The level of significance was 5%.ResultsImmature: No severe complications (PN, IRR, ARR, MBL, or TL) were diagnosed during follow up. Mature: Estimated risk after a 10-year follow up: PN: 56% (95% confidence interval (CI): 48.1–63.9), IRR: 2.5% (95% CI: 0–5.1), ARR: 2.1% (95% CI: 0.1–4.1), MBL: 2.4% (95% CI: 0.3–4.4), and TL: 7.8% (95% CI: 0–15.7). The following factors significantly increased the risk of PN in teeth with mature root development: fracture in relation to apex (hazard ratio (HR): 2.6 (95% CI: 0.2 – 5.7), P = 0.01), displacement in the horizontal part of the fracture >2 mm (HR: 1.8; 95% CI: 1.1–3.2, P = 0.03), incomplete repositioning (HR: 2.1 (95% CI: 1.3–3.5), P = 0.003), and age >30 years (HR: 2.3 (95% CI: 1.1–4.6), P = 0.02). The type of splint (rigid or flexible), the duration of splinting (more or less than 4 weeks), and the administration of antibiotics did not affect the risk of PN.Conclusion Teeth involved in alveolar process fractures appear, apart from PN, to have a good prognosis. A conservative treatment approach is recommended.
    Dental Traumatology 10/2015; DOI:10.1111/edt.12229 · 1.60 Impact Factor
  • Randi Grøn · Thomas A Gerds · Per K Andersen ·
    [Show abstract] [Hide abstract]
    ABSTRACT: Poisson regression is an important tool in register-based epidemiology where it is used to study the association between exposure variables and event rates. In this paper, we will discuss the situation with 'large n and small p', where n is the sample size and p is the number of available covariates. Specifically, we are concerned with modeling options when there are time-varying covariates that can have time-varying effects. One problem is that tests of the proportional hazards assumption, of no interactions between exposure and other observed variables, or of other modeling assumptions have large power due to the large sample size and will often indicate statistical significance even for numerically small deviations that are unimportant for the subject matter. Another problem is that information on important confounders may be unavailable. In practice, this situation may lead to simple working models that are then likely misspecified. To support and improve conclusions drawn from such models, we discuss methods for sensitivity analysis, for estimation of average exposure effects using aggregated data, and a semi-parametric bootstrap method to obtain robust standard errors. The methods are illustrated using data from the Danish national registries investigating the diabetes incidence for individuals treated with antipsychotics compared with the general unexposed population. Copyright © 2015 John Wiley & Sons, Ltd.
    Statistics in Medicine 10/2015; DOI:10.1002/sim.6755 · 1.83 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Background Biomarkers predicting response to primary androgen deprivation therapy (ADT) and risk of castration-resistant prostate cancer (CRPC) is lacking. We aimed to analyse the predictive value of ERG expression for development of CRPC.Methods In total, 194 patients with advanced and/or metastatic prostate cancer (PCa) treated with first-line castration-based ADT were included. ERG protein expression was analysed in diagnostic specimens using immunohistochemistry (anti-ERG, EPR3864). Time to CRPC was compared between ERG subgroups using multiple cause-specific Cox regression stratified on ERG-status. Risk reclassification and time-dependent area under the ROC curves were used to assess the discriminative ability of ERG-status. Time to PSA-nadir, proportion achieving PSA-nadir ≤0.2 ng/ml, and risk of PCa-specific death were secondary endpoints.ResultsMedian follow-up was 6.8 years (IQR: 4.9–7.3). In total, 105 patients (54.1%) were ERG-positive and 89 (45.9%) were ERG-negative. No difference in risk of CRPC was observed between ERG subgroups (P = 0.51). Median time to CRPC was 3.9 years (95%CI: 3.2–5.1) and 4.5 years (95%CI: 2.3-not reached) in the ERG-positive and ERG-negative group, respectively. Compared to a model omitting ERG-status, the ERG-stratified model showed comparable AUC values 1 year (77.6% vs. 78.0%, P = 0.82), 2 years (71.7% vs. 71.8%, P = 0.85), 5 years (68.5% vs. 69.9%, P = 0.32), and 8 years (67.9% vs. 71.4%, P = 0.21) from ADT initiation. No differences in secondary endpoints were observed.ConclusionsERG expression was not associated with risk of CRPC suggesting that ERG is not a candidate biomarker for predicting response to primary ADT in patients diagnosed with advanced and/or metastatic PCa. Prostate © 2015 Wiley Periodicals, Inc.
    The Prostate 05/2015; 75(14). DOI:10.1002/pros.23026 · 3.57 Impact Factor

  • European Urology Supplements 04/2015; 14(2):e529. DOI:10.1016/S1569-9056(15)60522-X · 3.37 Impact Factor

  • European Urology Supplements 04/2015; 14(2):e323. DOI:10.1016/S1569-9056(15)60320-7 · 3.37 Impact Factor
  • Michael W Kattan · Thomas A Gerds ·
    [Show abstract] [Hide abstract]
    ABSTRACT: Choosing to replace or maintain an existing cancer staging system is a difficult task. The system plays a critical role in patient counselling and treatment decision making because the staging system conveys prognosis. Many issues may be considered when deciding the preferred system (i.e. old or new), such as the level of evidence for one or more factors included in the system or the general opinions of expert clinicians. However, given the major objective of estimating prognosis on an ordinal scale, we argue that the rival staging system candidates should be compared on their ability to predict outcome. We sought to outline an algorithm that would compare two rival ordinal systems on their predictive ability. We devised an algorithm based largely on the concordance index, which is appropriate for comparing two models in their ability to rank observations. We demonstrate our algorithm with a prostate cancer staging system example. We have provided an algorithm for selecting the preferred staging system based on prognostic accuracy. It appears to be useful for the purpose of selecting between two ordinal prediction models. © The Author(s) 2015.
    Clinical Trials 02/2015; 12(4). DOI:10.1177/1740774515572614 · 1.93 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Background: Telehealth interventions for patients with chronic obstructive pulmonary disease (COPD) have focused primarily on stable outpatients. Telehealth designed to handle the acute exacerbation that normally requires hospitalization could also be of interest. The aim of this study was to compare the effect of home-based telehealth hospitalization with conventional hospitalization for exacerbation in severe COPD. Materials and methods: A two-center, noninferiority, randomized, controlled effectiveness trial was conducted between June 2010 and December 2011. Patients with severe COPD admitted because of exacerbation were randomized 1:1 either to home-based telehealth hospitalization or to continue standard treatment and care at the hospital. The primary outcome was treatment failure defined as re-admission due to exacerbation in COPD within 30 days after initial discharge. The noninferiority margin was set at 20% of the control group's risk of re-admission. Secondary outcomes were mortality, need for manual or mechanical ventilation or noninvasive ventilation, length of hospitalization, physiological parameters, health-related quality of life, user satisfaction, healthcare costs, and adverse events. Results: In total, 57 patients were randomized: 29 participants in the telehealth group and 28 participants in the control group. Testing the incidence of re-admission within 30 days after discharge could not confirm noninferiority (lower 95% confidence limit [CL], -24.8%; p=0.35). Results were also nonsignificant at 90 days (lower 95% CL, -16.2%; p=0.33) and 180 days (lower 95% CL, -16.6%; p =0.33) after discharge. Superiority testing on secondary outcomes showed nonsignificant differences between groups. Healthcare costs have not yet been evaluated. Conclusions: Whether home-based telehealth hospitalization is noninferior to conventional hospitalization requires further investigation. The results indicate that a subgroup of patients with severe COPD can be treated for acute exacerbation at home using telehealth, without the physical presence of health professionals and with a proper organizational "back-up."
    Telemedicine and e-Health 02/2015; 21(5). DOI:10.1089/tmj.2014.0098 · 1.67 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Objectives Maternal body mass index (BMI), birth weight, and preschool BMI may help identify children at high risk of overweight as they are (1) similarly linked to adolescent overweight at different stages of the obesity epidemic, (2) linked to adult obesity and metabolic alterations, and (3) easily obtainable in health examinations in young children. The aim was to develop early childhood prediction models of adolescent overweight, adult overweight, and adult obesity.Methods Prediction models at various ages in the Northern Finland Birth Cohort born in 1966 (NFBC1966) were developed. Internal validation was tested using a bootstrap design, and external validation was tested for the model predicting adolescent overweight using the Northern Finland Birth Cohort born in 1986 (NFBC1986).ResultsA prediction model developed in the NFBC1966 to predict adolescent overweight, applied to the NFBC1986, and aimed at labelling 10% as “at risk” on the basis of anthropometric information collected until 5 years of age showed that half of those at risk in fact did become overweight. This group constituted one-third of all who became overweight.Conclusions Our prediction model identified a subgroup of children at very high risk of becoming overweight, which may be valuable in public health settings dealing with obesity prevention.
    Obesity 01/2015; 23(1). DOI:10.1002/oby.20921 · 3.73 Impact Factor
  • Frank Eriksson · Thomas Alexander Gerds · Emmanuel Lesaffre ·
    [Show abstract] [Hide abstract]
    ABSTRACT: The marginal approach and the conditional approach are two different ways to model clustered dental failure time data. We compare the two approaches in the context of a Cox regression analysis, where the aim is to estimate the effect of a covariate (e.g., dental treatment) on the risk of failure. Specifically, we treat within-cluster correlation as if it was introduced by unobserved cluster level covariates, and study the small sample behaviour of the marginal and the conditional approach. We show that in a non-randomized setting where an unobserved cluster variable is correlated with the variable of interest, both the marginal and the conditional approaches can give misleading results. We argue that this is an important message, since most often it is assumed that the frailty term and the covariates of interest are independent.
    Statistical Modelling 12/2014; 14(6):549-566. DOI:10.1177/1471082X14535518 · 0.98 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Nonsteroidal anti-inflammatory drugs (NSAIDs) are assumed to increase bleeding risk, but their actual relation to serious bleeding in patients with atrial fibrillation (AF) who are receiving antithrombotic medication is unknown.
    Annals of internal medicine 11/2014; 161(10):690-8. DOI:10.7326/M13-1581 · 17.81 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Background Transit-time flow measurement (TTFM) is a commonly used intraoperative method for evaluation of coronary artery bypass graft (CABG) anastomoses. This study was undertaken to determine whether TTFM can also be used to predict graft patency at one year postsurgery. Methods Three hundred forty-five CABG patients with intraoperative graft flow measurements and one year angiographic follow-up were analyzed. Graft failure was defined as more than 50% stenosis including the string sign. Logistic regression analysis was used to analyze the risk of graft failure after one year based on graft vessel type, anastomatic configuration, and coronary artery size. ResultsNine hundred eighty-two coronary anastomoses were performed of which 12% had signs of graft failure at one year angiographic follow-up. In internal mammary arteries (IMAs), analysis showed a 4% decrease in graft failure odds for every 1mL/min increase in TTFM (OR=0.96, CI=[0.93; 0.99], p=0.005). ROC analysis showed good discriminative ability for TTFM alone AUC=69.5% in IMA grafts. For single-vein grafts the decrease in graft failure odds was 2% for every 1mL/min increase in TTFM (OR=0.98; CI=[0.97; 1.00], p=0.059) and AUC of 59.9%. There were no significant relationships between TTFM and graft failure in other graft types or graft configurations. Conclusion The TTFM method has good discriminative ability for assessing the risk of graft failure in certain graft types within the first year after CABG surgery and is a valuable instrument for intraoperative quality assessment of bypass grafts. doi: 10.1111/jocs.12471 (J Card Surg 2015;30:47-52)
    Journal of Cardiac Surgery 11/2014; 30(1). DOI:10.1111/jocs.12471 · 0.89 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Background Compelling biomarkers identifying prostate cancer patients with a high risk of progression during active surveillance (AS) are needed. Objective To examine the association between ERG expression at diagnosis and the risk of progression during AS. Design, setting, and participants This study included 265 patients followed on AS with prostate-specific antigen (PSA) measurements, clinical examinations, and 10–12 core rebiopsies from 2002 to 2012 in a prospectively maintained database. ERG immunohistochemical staining was performed on diagnostic paraffin-embedded formalin-fixed sections with a ready-to-use kit (anti-ERG, EPR3864). Men were characterised as ERG positive if a minimum of one tumour focus demonstrated ERG expression. Outcome measurements and statistical analysis Overall AS progression was defined as clinical progression: increased clinical tumour category ≥cT2b by digital rectal examination and ultrasound, and/or histopathologic progression: upgrade of Gleason score, more than three positive cores or bilateral positive cores, and/or PSA progression: PSA doubling time <3 yr. Risk of progression was analysed using multiple cause-specific Cox regression and stratified cumulative incidences (Aalen-Johansen method). Curatively intended treatment, watchful waiting, and death without progression were treated as competing events. Results and limitations A total of 121 of 142 ERG-negative and 96 of 123 ERG-positive patients had complete diagnostic information. In competing risk models, the ERG-positive group showed significantly higher incidences of overall AS progression (p < 0.0001) and of the subgroups PSA progression (p < 0.0001) and histopathologic progression (p < 0.0001). The 2-yr cumulative incidence of overall AS progression was 21.7% (95% confidence interval [CI], 14.3–29.1) in the ERG-negative group compared with 58.6% (95% CI, 48.7–68.5) in the ERG-positive group. ERG positivity was a significant predictor of overall AS progression in multiple Cox regression (hazard ratio: 2.45; 95% CI, 1.62–3.72; p < 0.0001). The main limitation of this study is its observational nature. Conclusions In our study, ERG positivity at diagnosis can be used to estimate the risk of progression during AS. If confirmed, ERG status can be used to individualise AS programmes. Patient summary The tissue biomarker ERG identifies active surveillance patients with an increased risk of disease progression.
    European Urology 11/2014; 66(5). DOI:10.1016/j.eururo.2014.02.058 · 13.94 Impact Factor
  • Jørgen Hilden · Thomas A Gerds ·
    [Show abstract] [Hide abstract]
    ABSTRACT: The 'integrated discrimination improvement' (IDI) and the 'net reclassification index' (NRI) are statistics proposed as measures of the incremental prognostic impact that a new biomarker will have when added to an existing prediction model for a binary outcome. By design, both measures were meant to be intuitively appropriate, and the IDI and NRI formulae do look intuitively plausible. Both have become increasingly popular. We shall argue, however, that their use is not always safe. If IDI and NRI are used to measure gain in prediction performance, then poorly calibrated models may appear advantageous, and in a simulation study, even the model that actually generates the data (and hence is the best possible model) can be improved on without adding measured information. We illustrate these shortcomings in actual cancer data as well as by Monte Carlo simulations. In these examples, we contrast IDI and NRI with the area under ROC and the Brier score. Unlike IDI and NRI, these traditional measures have the characteristic that prognostic performance cannot be accidentally or deliberately inflated. Copyright © 2013 John Wiley & Sons, Ltd.
    Statistics in Medicine 08/2014; 33(19). DOI:10.1002/sim.5804 · 1.83 Impact Factor
  • Thomas A. Gerds · Jørgen Hilden ·

    Statistics in Medicine 08/2014; 33(19). DOI:10.1002/sim.6212 · 1.83 Impact Factor
  • Source
    Margaret S. Pepe · Jing Fan · Ziding Feng · Thomas Gerds · Jorgen Hilden ·
    [Show abstract] [Hide abstract]
    ABSTRACT: The Net Reclassification Index (NRI) is a very popular measure for evaluating the improvement in prediction performance gained by adding a marker to a set of baseline predictors. However, the statistical properties of this novel measure have not been explored in depth. We demonstrate the alarming result that the NRI statistic calculated on a large test dataset using risk models derived from a training set is likely to be positive even when the new marker has no predictive information. A related theoretical example is provided in which an incorrect risk function that includes an uninformative marker is proven to erroneously yield a positive NRI. Some insight into this phenomenon is provided. Since large values for the NRI statistic may simply be due to use of poorly fitting risk models, we suggest caution in using the NRI as the basis for marker evaluation. Other measures of prediction performance improvement, such as measures derived from the receiver operating characteristic curve, the net benefit function, and the Brier score, cannot be large due to poorly fitting risk functions.
    Statistics in Biosciences 08/2014; 7(2). DOI:10.1007/s12561-014-9118-0
  • Thomas A Gerds · Per K Andersen · Michael W Kattan ·
    [Show abstract] [Hide abstract]
    ABSTRACT: A predicted risk of 17% can be called reliable if it can be expected that the event will occur to about 17 of 100 patients who all received a predicted risk of 17%. Statistical models can predict the absolute risk of an event such as cardiovascular death in the presence of competing risks such as death due to other causes. For personalized medicine and patient counseling, it is necessary to check that the model is calibrated in the sense that it provides reliable predictions for all subjects. There are three often encountered practical problems when the aim is to display or test if a risk prediction model is well calibrated. The first is lack of independent validation data, the second is right censoring, and the third is that when the risk scale is continuous, the estimation problem is as difficult as density estimation. To deal with these problems, we propose to estimate calibration curves for competing risks models based on jackknife pseudo-values that are combined with a nearest neighborhood smoother and a cross-validation approach to deal with all three problems. Copyright © 2014 John Wiley & Sons, Ltd.
    Statistics in Medicine 08/2014; 33(18). DOI:10.1002/sim.6152 · 1.83 Impact Factor
  • Source
    Jacob Pedersen · Thomas Alexander Gerds · Jakob Bue Bjorner · Karl Bang Christensen ·
    [Show abstract] [Hide abstract]
    ABSTRACT: Background Targeted interventions for the long-term sick-listed may prevent permanent exclusion from the labour force. We aimed to develop a prediction method for identifying high risk groups for continued or recurrent long-term sickness absence, unemployment, or disability among persons on long-term sick leave. Methods We obtained individual characteristics and follow-up data from the Danish Register of Sickness Absence Compensation Benefits and Social Transfer Payments (RSS) during 2004 to 2010 for 189,279 Danes who experienced a period of long-term sickness absence (4+ weeks). In a learning data set, statistical prediction methods were built using logistic regression and a discrete event simulation approach for a one year prediction horizon. Personalized risk profiles were obtained for five outcomes: employment, unemployment, recurrent sickness absence, continuous long-term sickness absence, and early retirement from the labour market. Predictor variables included gender, age, socio-economic position, job type, chronic disease status, history of sickness absence, and prior history of unemployment. Separate models were built for times of economic growth (2005–2007) and times of recession (2008–2010). The accuracy of the prediction models was assessed with analyses of Receiver Operating Characteristic (ROC) curves and the Brier score in an independent validation data set. Results In comparison with a null model which ignored the predictor variables, logistic regression achieved only moderate prediction accuracy for the five outcome states. Results obtained with discrete event simulation were comparable with logistic regression. Conclusions Only moderate prediction accuracy could be achieved using the selected information from the Danish register RSS. Other variables need to be included in order to establish a prediction method which provides more accurate risk profiles for long-term sick-listed persons.
    BMC Public Health 05/2014; 14(1):494. DOI:10.1186/1471-2458-14-494 · 2.26 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: We introduce a new approach to competing risks using random forests. Our method is fully non-parametric and can be used for selecting event-specific variables and for estimating the cumulative incidence function. We show that the method is highly effective for both prediction and variable selection in high-dimensional problems and in settings such as HIV/AIDS that involve many competing risks.
    Biostatistics 04/2014; 15(4). DOI:10.1093/biostatistics/kxu010 · 2.65 Impact Factor

Publication Stats

2k Citations
382.96 Total Impact Points


  • 2009-2015
    • IT University of Copenhagen
      København, Capital Region, Denmark
    • Herlev Hospital
      • Department of Pathology
      Herlev, Capital Region, Denmark
  • 2012
    • Catholic University of Louvain
      Лувен-ла-Нев, Walloon, Belgium
  • 2008
    • New York University College of Dentistry
      New York City, New York, United States
  • 2003-2008
    • University of Freiburg
      • Institute of Medical Biometry and Medical Informatics
      Freiburg, Baden-Württemberg, Germany
  • 2005
    • Universitätsklinikum Freiburg
      • Department of Prosthodontics
      Freiburg an der Elbe, Lower Saxony, Germany
  • 2004
    • Deutsche Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e.V.
      Freiburg, Baden-Württemberg, Germany